How Conflux works, and how much to trust it

Conflux watches sensors across the Northern Territory and tells you when something looks unusual — in plain English. Here is exactly how it decides that, where the readings come from, and what it is not able to do.

How fresh is this?

The whole system, in four steps

Sensors report

Automatic weather stations across the Territory send readings on their own schedules, and a marine buoy off Darwin reports the water as well as the air above it.

We line them up

Air and sea measure on different clocks, so everything is resampled onto one hourly timeline. Only then can a weather mast and a buoy be compared directly.

Four checks look for the unusual

Each one has a different speciality, and each is measured against a station's own history rather than against an absolute number. What counts as normal in Darwin is not normal in Alice Springs.

We write it down in plain English

Anything flagged is rated the same three ways — how many checks agree, how far outside normal, and how long it lasted — and turned into a sentence you can act on.

Where the readings come from

Bureau of Meteorology

Automatic weather stations: air temperature, humidity, pressure and wind. Used under BOM's research terms — Conflux publishes what it concludes, never the raw observations.

Australian Institute of Marine Science

The Darwin National Reference Station buoy, which reports water and air temperature from the same platform. That co-location is what makes the air–sea comparison possible at all.

How we decide how confident to be

Three things, every time. How many of the checks agree — they read overlapping data, so agreement supports a finding rather than proving it. How far outside normal the readings sit, measured against that station's own history. How long it lasted, because one strange hour is usually noise and eleven is not.

Looking normal

Nothing flagged at the latest reading.

Worth a look

Flagged, but brief, mild, or by a single check.

Needs attention

Two or more checks agree, well outside normal, and sustained.

What Conflux does not do

It does not forecast. Everything here describes readings that have already arrived. Conflux cannot tell you about tomorrow, or about anywhere a sensor is not.

It is not an official warning service. For cyclones, floods and fire, the Bureau of Meteorology and the NT Emergency Service are the authorities. Conflux is a second pair of eyes, not a siren.

It does not know why something is unusual. A flag can be real weather or a failing sensor, and Conflux cannot tell the two apart. Deciding which it is remains a job for a person.

It does not cover everywhere. A few dozen stations, and exactly one place — Darwin — where air and sea can be compared. Large parts of the Territory have no sensor anywhere near them.

It has a short memory. The store is rebuilt each run from rolling windows of roughly three days, so “normal” is measured against a thin baseline. Longer retention is the next piece of work.

Its air–sea check is not yet proven. That check flags far more often than the other three. We have tested it against clean synthetic data and believe it is responding to a real daily rhythm rather than misfiring — but until it has been scored against a set of confirmed events, treat its flags as a prompt to look, not as a conclusion.

Still unsure about something on a card? Every finding on the briefing opens up the same way this page does — the reasoning, then the stations, then the models.